Get right probability for Classify Colors in MMA 11.3.0.0 on RPi?

Questions
, ,
1

This seems to execute incorrectly on RPi with incorrect probability any thoughts would be appreciated! Thanks.

Directly from the Wolfram Documentation Center :

Classify >> Examples >> Scope >> Custom Classifiers

Train a classifier on image examples, gathered by their class:

probabiltyOfClassifyColor.jpg
probabiltyOfClassifyColor.jpg783×340 28 KB

2

Dear Terrance, this behavior is strange because the automation is supposed to discard models that give bad probabilities like this, but it can happen in small dataset. Most likely the “RandomForest” method is selected here (you can obtain such information by doing ClassifierInformation[c] ), and this method is not great to give good probabilities.

You can try to improve the model quality by doing

Classify[..., PerformanceGoal ->"Quality"]

or setting up a time goal, e.g.

Classify[..., TimeGoal -> 10]

or choosing a method yourself, e.g.

Classify[..., Method->"LogisticRegression"]

Finally, you can try to add more data, this should help the automation to do the right thing by testing on larger test sets.

Thanks,
Etienne